Guide · Governance
AI Production Readiness
AI production readiness is the engineering question of whether a specific AI application can safely operate in production—security, evaluation, operability, cost, and governance—not whether a demo worked.
AI production readiness is what APRF specifies with domains, checks, and gated profiles. Canonical framework: AI Production Readiness Framework
Definition
AI production readiness means a specific AI application (chatbot, agent, RAG system, MCP server, voice assistant, and similar) can safely operate in production under real users, real attackers, and real bills.
It is not:
- A model leaderboard score
- A single “readiness percentage” badge
- Proof that you are compliant with NIST, ISO, or SOC 2
- The same as broad AI readiness (organizational preparedness)
What production breaks that demos hide
Production AI fails across domains demos skip: prompt injection and tool abuse, leaked keys, unversioned prompts, missing evaluation gates, unbounded spend, weak human approval for high-impact actions, and no incident playbook for model or agent failures.
A credible AI production readiness assessment is therefore gated: mandatory checks either pass (with evidence) or block release—not average into a green score.
How the AI Production Readiness Framework specifies it
The AI Production Readiness Framework (APRF) is a vendor-neutral working draft that defines:
- Eight engineering domains and focused pillars
- Checks with required artifacts and pass conditions
- Profiles (Core and Regulated) that select mandatory gates by criticality
- Lenses that add gates for system types (RAG, agents, voice, coding agents)
- Versioning, RFC process, and conformance citation rules
Machine-readable catalog: /aprf/spec/. Human depth: /aprf/how/.
Practical path
1. Identify the system’s criticality tier and applicable lenses.
2. Run the Core Profile gates for customer-facing systems; use Regulated for Tier‑3 / mission-critical.
3. Treat self-attestation as a starting map; evidence-backed assessment is required for serious conformance claims.
4. Cite APRF version, profile, capability level, and open blockers in any public claim.
Related practice guides
APRF practice guides cover concrete failure modes—secrets in the frontend, AI API bill spikes, evaluation gates, agent autonomy limits—and map back to pillars. Start from the guides index or the production readiness overview.
Next: AI Production Readiness Framework
Open the related pillar specification for mandatory checks, artifacts, and pass conditions. Self-attest is optional.
Related
Frequently asked questions
- What is AI production readiness?
- AI production readiness means a specific AI application can safely operate in production: measurable controls for security, evaluation, operability, cost, resilience, and governance—not a demo or a vanity percentage score.
- How do you measure AI production readiness?
- Use gated mandatory checks with artifacts and pass conditions against a pinned profile (for example APRF Core or Regulated). Cite version, criticality, capability level, profile, and any open blockers. Avoid single percentage badges.
- What is the AI Production Readiness Framework?
- The AI Production Readiness Framework (APRF) is a vendor-neutral working-draft specification for AI production readiness: eight domains, measurable checks, gated profiles, and an open RFC stewardship path. StackRail publishes the working draft.